MovieLens Recommender Algorithm Comparison

This project compares the performance of five collaborative filtering algorithms: KNNBasic, SVD, NMF, SlopeOne, and CoClustering. Using the MovieLens ml-latest-small dataset (100,836 ratings), each model was implemented with the Surprise library and evaluated using 5-fold cross-validation. The goal is to measure accuracy (RMSE, MAE), training time, and prediction time to determine which approach performs best for recommendation tasks.

Dataset

MovieLens ml-latest-small

• 100,836 ratings
• 610 users
• 9,742 movies
• Rating scale: 0.5 to 5.0

Source: GroupLens Research
https://grouplens.org/datasets/movielens/

Algorithms Compared

• KNNBasic: Neighborhood-based collaborative filtering
• SVD: Matrix factorization
• NMF: Non-negative matrix factorization
• SlopeOne: Item-based differences algorithm
• CoClustering: User–item co-clustering

Methodology

All algorithms were evaluated using:

• 5-fold cross-validation
• RMSE
• MAE
• Fit time
• Test time

All experiments were conducted in Python using the Surprise library.

Summary of Results

SVD achieved the lowest RMSE and MAE and provided the best balance of accuracy and speed. Other algorithms performed as follows:

Algorithm	RMSE	MAE	Notes
SVD	~0.87	~0.67	Best overall performance
SlopeOne	~0.90	~0.69	Moderate accuracy, very slow
NMF	~0.92	~0.71	Good accuracy, slower training
CoClustering	~0.95	~0.73	Weak performance
KNNBasic	~0.95	~0.73	Baseline performance

Project Structure

.
├── movielens/
│   ├── ratings.csv
│   ├── movies.csv
│   ├── tags.csv
│   ├── links.csv
│   └── README.txt
├── notebooks/
│   └── discussion19_1.ipynb
├── README.md
└── requirements.txt

## Running the Project

Install dependencies:



pip install -r requirements.txt


Open the notebook:



jupyter notebook notebooks/discussion19_1.ipynb


## Requirements

- Python 3.8 or higher  
- pandas  
- numpy  
- scikit-surprise  
- jupyter notebook  

## License

This project uses the MovieLens dataset under the GroupLens Research terms of use.